This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [m32c]Problem in bit instructions generation
- From: Roger Sayle <roger at eyesopen dot com>
- To: DJ Delorie <dj at redhat dot com>, "Naveen H.S." <naveenh at KPITCummins dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Mon, 10 Jul 2006 19:36:07 -0600 (MDT)
- Subject: Re: [m32c]Problem in bit instructions generation
Hi DJ and Naveen,
> The trick is to figure out what kinds of rtl patterns gcc is looking
> for, in terms of operators that gcc *does* know about (like andqi3 or
> cmp/jmp). What I usually do is edit insn-recog.c (in the build
> directory) to debug_rtx() each pattern gcc tries to recognize. Then
> you adjust your .md file to look like what gcc is looking for.
I'll admit to doing something very similar, but instead of inserting
code into the machine-generated "recog", I prefer to place the
debug_rtx in combine.c:recog_for_combine. One possibility, is that
GCC is managing to generate suitable zero_extract instructions but
the target's rtx_costs indicate that this instruction is more expensive
than the primitive bit manipulation operations it was synthesized from.
Roger
--